Machine Troubleshooting AI Assistant

An advanced AI-powered chatbot system for industrial machine diagnostics and troubleshooting, built with Mistral 7B Instruct v0.3 and semantic search capabilities.

Features

AI-Powered Diagnostics

• Mistral 7B Model: Uses Mistral-7B-Instruct-v0.3 for intelligent troubleshooting responses
• RAG (Retrieval-Augmented Generation): Combines semantic search with AI generation for accurate solutions
• Context-Aware: Leverages historical maintenance data to provide relevant solutions

Semantic Search

• Vector Embeddings: Uses Sentence-BERT (all-MiniLM-L6-v2) for intelligent case matching
• Hybrid Search: Combines keyword matching and cosine similarity for optimal results
• Case History: Searches through past maintenance records to find similar issues

Conversational Interface

• Real-time Chat: Natural language interaction with the AI assistant
• Structured Diagnosis: Form-based troubleshooting for detailed analysis
• Multi-turn Conversations: Maintains context throughout the discussion

Analytics & Feedback

• User Feedback System: Collects ratings and comments to improve accuracy
• Statistics Dashboard: Tracks queries, success rates, and user contributions
• Export Functionality: Download chat history and correction submissions

Production Ready

• Optimized Performance: FP16 precision for efficient GPU usage (~13.5GB VRAM)
• Error Handling: Robust fallback mechanisms for reliability
• Scalable Architecture: Modular design for easy maintenance and updates

System Requirements

Hardware

• GPU: NVIDIA GPU with 16GB+ VRAM (tested on RTX 4090 20GB)
• RAM: 16GB+ system memory
• Storage: 100GB+ free space (for model downloads)

Software

• OS: Windows 10/11, Linux (Ubuntu 20.04+), or macOS
• Python: 3.8 or higher
• CUDA: 11.7+ (for GPU acceleration)

Installation

1. Clone the Repository

git clone https://github.com/Romilagarwal/new_chatbot.git
cd new_chatbot

2. Create Virtual Environment

python -m venv venv

# Windows
venv\Scripts\activate

# Linux/Mac
source venv/bin/activate

3. Install Dependencies

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip install transformers accelerate bitsandbytes
pip install flask flask-compress sentence-transformers
pip install pandas numpy scikit-learn python-dotenv pytz

4. Setup Environment Variables

Create a .env file in the project root:

# Optional: HuggingFace token for gated models
HF_TOKEN=your_huggingface_token_here

# Device configuration
DEVICE=cuda

# Dataset path
DATABASE_PATH=mix_dataset_final.csv

# Cache directory (optional)
HF_HOME=D:\.cache\huggingface\hub

5. Prepare Dataset

Place your machine maintenance dataset CSV file (mix_dataset_final.csv) in the project root with the following columns:

• Machine Type
• MACHINE (Machine Name)
• Problem Description
• Root Cause
• Action Taken

Usage

Starting the Application

python app.py

The application will:

1. Load the Mistral 7B model (~2-3 minutes first time)
2. Create embeddings for the reference dataset
3. Start the Flask server at http://172.19.66.141:9352
4. Automatically open your web browser

Using the Chatbot

Conversational Chat

Simply type your question in the chat interface:

"APMT machine showing alarm 203"
"How to fix welding spot machine not working?"
"Printer paper jam issue"

Structured Diagnosis

1. Select Machine Type from dropdown
2. Enter Machine Name (e.g., PRE-APMT, WELDING SPOT)
3. Describe the Problem in detail
4. Click Diagnose Problem

The system will:

• Find similar cases from the database
• Analyze patterns using AI
• Provide root cause analysis
• Suggest immediate actions
• Recommend preventive measures

Contributing Solutions

Help improve the system by submitting your solved cases:

1. Click Add Solution button
2. Fill in the machine details and solution
3. Submit for review

Project Structure

new_chatbot/
├── app.py                      # Main Flask application
├── advanced_model.py           # Mistral model wrapper
├── model_utils.py             # RAG and semantic search utilities
├── model_loader.py            # Model testing and verification
├── templates/
│   └── index.html             # Frontend interface
├── data/                      # Generated data directory
│   ├── feedback.csv           # User feedback
│   ├── user_corrections.csv   # Submitted solutions
│   └── chat_logs.json         # Chat history
├── mix_dataset_final.csv      # Your maintenance dataset
├── .env                       # Environment configuration
└── README.md                  # This file

Configuration

Model Settings

Mistral 7B Instruct v0.3 configuration in advanced_model.py:

• Precision: FP16 (float16)
• VRAM Usage: ~13.5GB
• Context Length: 32,768 tokens
• Temperature: 0.7 (troubleshooting), 0.8 (chat)
• Top-p: 0.9-0.92

Search Parameters

Adjust in model_utils.py:

threshold = 0.60        # Similarity threshold
top_n = 3              # Number of primary results
extra_results = 15     # Additional dropdown candidates

Server Settings

Modify in app.py:

host = '172.19.66.141'  # Server IP
port = 9352             # Server port
debug = False           # Production mode

API Endpoints

Endpoint	Method	Description
/	GET	Main application interface
/health	GET	System health check
/stats	GET	Usage statistics
/chat	POST	Conversational chat endpoint
/diagnose	POST	Structured diagnosis
/feedback	POST	Submit user feedback
/submit-correction	POST	Submit new solutions
/export/chat-history	GET	Export chat logs
/export/corrections	GET	Export user contributions

Performance

Model Performance

• First Load: 2-3 minutes (downloads ~87GB)
• Subsequent Loads: <1 minute (from cache)
• Generation Speed: 15-25 tokens/second
• Latency: Low (full GPU, no CPU offload)

Search Performance

• Embedding Creation: Batched processing for efficiency
• Query Time: <1 second for semantic search
• Accuracy: Hybrid approach ensures relevant results

Contributing

Contributions are welcome! Here's how you can help:

1. Report Issues: Found a bug? Open an issue
2. Submit Solutions: Use the built-in correction form
3. Code Improvements: Fork, improve, and submit PRs
4. Documentation: Help improve this README

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Mistral AI for the Mistral 7B model
• HuggingFace for Transformers library
• Sentence-Transformers for embedding models
• Flask team for the web framework

Contact

Romil Agarwal

• GitHub: @Romilagarwal
• Repository: new_chatbot

Troubleshooting

Model Not Loading

# Check GPU availability
python -c "import torch; print(torch.cuda.is_available())"

# Check disk space
# Ensure ~100GB free in cache directory

Out of Memory

• Reduce max_tokens in generation config
• Close other GPU applications
• Consider using quantization (4-bit) if needed

Slow Performance

• Verify CUDA is properly installed
• Check GPU utilization: nvidia-smi
• Ensure model is on GPU, not CPU

Dataset Issues

• Verify CSV format matches expected columns
• Check for missing values in critical columns
• Ensure proper encoding (UTF-8)

Future Enhancements

• Multi-language support
• Voice input/output
• Mobile app version
• Advanced analytics dashboard
• Fine-tuning on domain-specific data
• Integration with maintenance management systems
• Real-time notification system
• Multi-user role management

---

Built with ❤️ for industrial maintenance teams

Star ⭐ this repository if you find it helpful!